Application of Neural-Networks Designed on Approximate Reasoning Architecture to the Adjustment of VTR Tape Running Mechanisms(Journal of Japan Society for Fuzzy Theory and Systems)

概要

論文の詳細を見る
This paper describes the application of NARA (Neural-networks designed on Approximate Reasoning Architecture) to a concrete task and its evaluation. It is important that the system should have both of logic and learning function to realize the adaptation with keeping the original safety of the system. The NARA has these function. The NARA and an ordinary single NN (artificial neural networks) are trained using input/output data of the rule-based system that adjusts VTR tape-running mechanisms, and the performance of them is compared. The effect has shown that the error rate could be reduced from 2.2% to 1.4% for training data and from 4.6% to 3.9% for untraining data by introduction of the NARA instead of an ordinary single NN. Furthermore we have evaluated the introduction of fuzzy logic to NN construction, analyzed the tendency of error, and discussed the comparison between the NARA and current running system in factory. Finally significance of this paper in realizing the adaptation and future research directions are described.
日本知能情報ファジィ学会の論文
1991-11-15